SPEED CONTROL OF DC MOTOR

USING NOVEL NEURAL

NETWORK CONFIGURATION

A PROJECT THESIS SUBMITTED IN THE FUFILLMENT OF THE

REQUIREMENTS FOR THE DEGREE OF

BACHELOR OF TECHNOLOGY
IN
ELECTRICAL ENGINEERING
BY
MANISH MISHRA
ROLL NO. 10502014

DEPARTMENT OF ELECTRICAL ENGINEERING

NATIONAL INSTITUTE OF TECHNOLOGY
ROURKELA

2008-2009
SPEED CONTROL OF DC MOTOR
USING NOVEL NEURAL
NETWORK CONFIGURATION

A PROJECT THESIS SUBMITTED IN THE FUFILLMENT OF THE

REQUIREMENTS FOR THE DEGREE OF
BACHELOR OF TECHNOLOGY
IN
ELECTRICAL ENGINEERING
BY
MANISH MISHRA
ROLL NO. 10502014

UNDER THE GUIDANCE OF

PROF. J.K SATHPATHY

DEPARTMENT OF ELECTRICAL ENGINEERING

NATIONAL INSTITUTE OF TECHNOLOGY
ROURKELA

2008-2009
 ACKNOWLEDGEMENT

I am deeply indebted to Professor J.K SATHPATHY for his able guidance, expert advice and

showing me the right way of carrying out the research project. I am also thankful to him for

giving me the special attention and cooperation in between his very busy hours.

I am also grateful to our entire electrical faculty for their efforts to strengthen the

fundamental concepts so that one can take up any assignment of serious gravity with

confidence. I would like to take this opportunity to express my heart-felt indebtedness

towards Prof. P. C. Panda and Prof. B. D. Subudhi for their support and enlightenment.

At this moment I would also like to express my gratitude for the technical staff of our

laboratories. They have always helped me in every-way they can during my project work.

I also duly acknowledge the work of the people listed in references.

Date: MANISH MISHRA

(ROLL NO.10502014)
 NATIONAL INSTITUTE OF TECHNOLOGY ROURKELA

CERTIFICATE

This is to certify that the progress report of the thesis entitled, SPEED CONTROL OF DC

MOTOR USING NOVEL NEURAL NETWORK CONFIGURATION submitted by Shri

Manish Mishra in partial fulfilment of the requirements for the award of Bachelor of

Technology degree in Electrical Engineering at the National Institute of Technology

Rourkela (Deemed University), is an authentic work carried out by him under my supervision

and guidance.

To the best of my knowledge the matter embodied in the thesis has not been submitted to any

other University/Institute for the award of any degree or diploma.

Prof. J.K SATHPATHY

Date: Department of Electrical Engineering

Place: National Institute of Technology Rourkela

 CONTENTS

ITEMS TITLE PAGE NO.

1 LIST OF FIGURES 1

2 ABSTRACT 3

CHAPTER: 1 INTRODUCTION 4

CHAPTER: 2 BACKGROUND INFORMATION 8

2.1 Overview 9

2.2 Transfer Function 10

2.3 Three-Layer Network 11

2.4 Universal Approximators 11

2.5 Back propagation Algorithm 12

2.5.1 Steepest Descent Algorithm For The Approx. Mean 13

Square Error

CHAPTER: 3 SEPARATELY EXCITED DC MOTOR 14

3.1 Operation 15

3.1.1 Field And Armature Equations 15

3.1.2 Basic Torque Equation 16

3.2 Steady State Operation 17

3.2.1 Steady-State Torque And Speed 18

3.2.2 Torque And Speed Control 18

3.2.3 Variable Speed Operation 19

3.2.4 Base Speed And Field-Weakening 19

CHAPTER: 4 OBJECTIVE 23
 4.1 System Dynamics 24

4.2 Control System Applications 24

4.3 System Identification 25

4.4 Control Design 26

4.5 Discrete Time DC Motor Model 26

4.6 Identification Of DC Motor Model 27

4.7 Trajectory Control Of DC Motor Using Trained ANN 27

CHAPTER: 5 PROBLEM STATEMENT 29

CHAPTER: 6 GENERATION OF THE SIGNAL 32

6.1 Overview 33

6.2 Result 34

CHAPTER: 7 SYSTEM IDENTIFICATION 35

7.1 Overview 36

7.2 Block Diagram 38

7.3 C-Code For Training With Back-Propagation Of The 38

ANN

CHAPTER: 8 TRAJECTORY CONTROL 49

8.1 Overview 50

8.2 Plots 51

8.3 Overall Controller Structure 52

CHAPTER: 9 RESULTS 53

9.1 Table Of Outputs For Comparison 54

9.2 Plot Of Trained Ann From System Identification 66

Coding
 CHAPTER: 10 CONCLUSION 67

10.1 Discussions 68

10.2 Future Scope 68

CHAPTER: 11 BIBLIOGRAPHY 70

 LIST OF FIGURES

FIGURE TITLE PAGE

NO. NO.

Figure. 1 Basic Neural Network 9

Figure. 2 Symmetric Hard-limit Transfer Function 10

Figure. 3 Linear Transfer Function 10

Figure. 4 Log-Sigmoid Transfer Function 10

Figure. 5 Three-Layer Network 11

Figure. 6 Universal Approximators 11

Figure. 7 Separately Excited DC Motor 15

Figure. 8 Separately Excited DC Motor In Steady State 17

Figure. 9 Torque Vs Speed Characteristic For Different Armature Voltages 19

Figure. 10 Torque Vs Speed And Power Vs Speed Characteristic Of 19

Separately Excited DC Motor

Figure. 11 Typical Operating Regions Of Separately Excited DC Machines 20

Figure. 12 Steady-State Torque-Speed Operating Characteristics Of 21

Separately Excited DC Motors

Figure. 13 Torque-Speed Characteristic Curves For If=Const 21

Figure. 14 Plant Identification 25

Figure. 15 Neural Network Plant Model 25

Figure. 16 Model Reference Control Architecture 26

Figure. 17 Structure Of The ANN Identifier 27

Figure. 18 Actual And Estimated Rotor Speeds 34

1
Figure. 19 Three-Layer Feed forward ANN 36

Figure. 20 Structure Of The ANN For Identification Of The DC Motor 38

Figure 21 Actual And Estimated Terminal Voltage Of The DC Motor 52

Figure. 22 Tracking Performance For A Sinusoidal Reference Track 52

Figure. 23 The Overall Structure Of The Controller In Topology 53

Figure. 24 Trend Of Root-Mean-Squared (RMS) Error For The Entire 66

Training Set Over Several Training Iterations

2
 ABSTRACT

This paper uses Artificial Neural Networks (ANNs) in estimating speed and controlling it for

a separately excited DC motor. The rotor speed of the dc motor can be made to follow an

arbitrarily selected trajectory. The purpose is to achieve accurate trajectory control of the

speed, especially when the motor and load parameters are unknown.

Such a neural control scheme consists of two parts. One is the neural identifier which is used

to estimate the motor speed. The other is the neural controller which is used to generate a

control signal for a converter. These two neural networks are trained by Levenberg-

Marquardt back-propagation algorithm. ANNs used in this are the standard three layers feed-

forward neural network with sigmoid activation functions in the input and hidden layers

while linear activation function is employed for the output layer.

The conventional constant gain feedback controller fails to maintain the performance of the

system at acceptable levels under unknown dynamics in load torque. On the other hand,

ANNs act as an effective tool to implement the model and adaptive control in a complicated

non-linear system having expansive allocations.

The adaptive learning algorithm is formed in such a way that the learning rate is as large as

possible while maintaining the stability of the learning process. This simplifies the learning

process in terms of computation time, which is of special importance in real-time

implementation.

3
 Chapter 1

INTRODUCTION

4
The development of high performance motor drives is very important in industrial

applications. Generally, a high performance motor drive system must have good dynamic

speed command tracking and load regulating response.

D.C motors have long been the primary means of electric traction. D.C motor is considered a

SISO system having torque/speed characteristics compatible with most mechanical loads.

This makes a D.C motor controllable over a wide range of speeds by proper adjustment of its

terminal voltage. Recently, brushless D.C motors, induction motors, and synchronous motors

have gained widespread use in electric traction. However, there is a persistent effort towards

making them behave like dc motors through innovative design and control strategies. Hence

dc motors are always a good proving ground for advanced control algorithm because the

theory is extendable to other types of motors.

Many practical control issues (motor control problems):

Variable and unpredictable inputs

Noise propagation along a series of unit processes

Unknown parameters

Changes in load dynamics

Under these conditions, the conventional constant gain feedback controller fails to maintain

the performance of the system at acceptable levels. The incorporation of feed forward in

artificial neural networks is important for several reasons the dynamical properties of the

system, and in practice it may improve the performance. They are generally present in most

non-linear dynamical system and can be used to implement specific structures.

5
Advantages of using ANNs:

Learning ability

Massive parallelism

Fast adaptation

Inherent approximation capability

High degree of tolerance

Speed control techniques in separately excited dc motor:

Varying the armature voltage in the constant torque region.

In the constant power region, field flux should be reduced to achieve speed above the

rated speed.

Methods of speed control:

Traditionally rheostat armature control method was used for low power dc motors.

Use of conventional PID controllers.

Neural network controllers (NNC).

Constant power field weakening controller based on load-adaptive multi-input multi-

output linearization technique (in high speed regimes).

A single phase uniform PWM ac-dc buck-boost converter with only one switching

device used for armature voltage control.

Use of NARMA-L2 (Non-linear Auto-regressive Moving Average) controller in the

constant torque region.

Through experience gained in designing trajectory controllers based on self-tuning and PID

control, it is seen that the neural network controller gives comparable performance in speed

tracking. In addition to those mentioned above, a unique advantage of the neural network

6
controller is its ability to cope with bad measurement data that occur during training and

testing. However, a key drawback is the inadequate integral gain in the feedback loop,

resulting in steady-state errors of the shaft position. Direct position tracking can alleviate this

problem.

The traditional means adapted by the motion control industry with motor drives has been the

approach of linearizing the system dynamics and designing a linear feedback controller.

However, in high-performance motor drives such an approach is seldom satisfactory, as it

results in poor speed and position tracking when sudden changes in load result in continuous

acceleration or deceleration of the motor/load system. Adaptation is necessary to ensure

optimal performance.

7
 Chapter 2

BACKGROUND INFORMATION

8
2.1 OVERVIEW:

The neural network consists of junctions which are connected with LINKS, also called

processing units. For each junction a number is ordered, this number is called weight. The

weights are the tools for the long distance information storing in the neural network, the

learning process occurring with the appropriate modification of weights. These weights are

modified so that the network input/output behaviour is in consonance with the environment,

which provide the input data.

The calculation algorithm consists of two basic steps:

Calculation of the output of the network, with inputs and weights.

Modification of weights with learning algorithm.

A single input neuron consists of a scalar input p multiplied by the scalar weight w to

form wp which is fed to the summer along with bias b multiplied by 1.

Figure 1: Basic Neural Network

The net input is wp+b and the output a is;

a=f (WP+b);

f- Transfer function

W & b can be adjusted by learning rule.

9
2.2 TRANSFER FUNCTION:

HARD-LIMIT TF:

a = hardlim (n)

Figure 2: Symmetric Hard-limit Transfer Function

LINEAR TF:

a = purelin (n)

Figure 3: Linear Transfer Function

LOG-SIGMOID TF:

a = logsig (n)

i.e. a = 1/ (1+exp (-n));

Figure 4: Log-Sigmoid Transfer Function

Used in multilayer n/w and trained using BPA as it is differentiable.

10
2.3 THREE-LAYER NETWORK:

Figure 5: Three-Layer Network

3rd layer - OUTPUT LAYER

1st & 2nd layer HIDDEN LAYER

2.4 UNIVERSAL APPROXIMATORS:

Figure 6: Universal Approximators

11
2.5 BACK PROPAGATION ALGORITHM:

Procedure for selecting parameters- TRAINING the n/w.

BPA is a method of TRAINING.

Based on gradient descent.

For a multilayer n/w, BPA is a gradient descent optimization procedure where we

minimize a mean square error called PERFORMANCE INDEX (PI).

Pq input to n/w

Aq n/w output

Tq corresponding target output

eq = Tq-Aq

X Vector containing all n/w w & b.

Q- Total number of Errors in the network

F(x)- Performance Index

For Multiple outputs;

Using stochastic approx.;

Where the expectation of the squared error has been replaced by the squared error at iteration

k.

12
2.5.1 STEEPEST DESCENT ALGORITHM FOR THE APPROX. MEAN SQUARE

ERROR:

i, j, m- Layers of neurons

F- Perfromance Index

w, b- respective weights and bias as per their sub-script and superscript.

13
 Chapter 3

SEPARATELY EXCITED DC MOTOR

14
 Figure 7: Separately Excited DC Motor

The field windings are used to excite the field flux.

Armature current is supplied to the rotor via brush and commutator for the mechanical

work.

Interaction of field flux and armature current in the rotor produces torque.

3.1 OPERATION:

When a separately excited motor is excited by a field current of if and an armature current of

ia flows in the circuit, the motor develops a back emf and a torque to balance the load torque

at a particular speed.

The if is independent of the ia. Each windings are supplied separately. Any change in the

armature current has no effect on the field current.

The if is normally much less than the ia.

3.1.1 FIELD AND ARMATURE EQUATIONS:

Instantaneous field current:

15
3.1.2 BASIC TORQUE EQUATION:

16
3.2 STEADY STATE OPERATION:

Figure 8: Separately Excited DC Motor In Steady State

17
3.2.1 STEADY-STATE TORQUE AND SPEED:

3.2.2 TORQUE AND SPEED CONTROL:

From the derivation, several important facts can be deduced for steady-state operation of

DC motor.

For a fixed field current, or flux (If), the torque demand can be satisfied by varying the

armature current (Ia).

The motor speed can be varied by:

controlling Va (voltage control)

controlling Vf (field control)

These observations lead to the application of variable DC voltage for controlling the speed

and torque of DC motor.

18
3.2.3 VARIABLE SPEED OPERATION:

Figure 9: Torque Vs Speed Characteristic For Different Armature Voltages

Family of steady-state torque speed curves for a range of armature voltage can be drawn as

above.

The speed of DC motor can simply be set by applying the correct voltage.

Note that speed variation from no-load to full load (rated) can be quite small. It depends on

the armature resistance.

3.2.4 BASE SPEED AND FIELD-WEAKENING:

Figure 10: Torque Vs Speed And Power Vs Speed Characteristic Of Separately Excited DC Motor

Base speed: (wbase)

the speed which correspond to the rated Va, rated Ia and rated If.

19
 Constant Torque region (w > wbase)

Ia and If are maintained constant to met torque demand. Va is varied to control the speed.

Power increases with speed.

Constant Power region (w > wbase)

Va is maintained at the rated value and if is reduced to increase speed. However, the power

developed by the motor (= torque x speed) remains constant. This phenomenon is known as

field weakening.

Figure 11: Typical Operating Regions Of Separately Excited DC Machines

20
 Figure 12: Steady-State Torque-Speed Operating Characteristics Of Separately Excited DC Motors

Figure 13: Torque-Speed Characteristic Curves For If=Const.

A high performance drive system consists of a motor, a converter, and a controller integrated

to perform a precise mechanical manoeuvre. This requires the shaft speed and/or the motor to

21
closely follow a specified trajectory regardless of unknown load variations and other

parameter uncertainties. One way to compensate for these uncertainties is to identify the

motor/load dynamics through the parameters of a pre-defined model. These model parameters

can then be manipulated using different control strategies to come up with the controller

design.

If the unknown dynamics are non-stationary, the identification has to be performed online.

However, if the dynamics are stationary, offline identification is adequate.

e.g. in self-tuning control, motor/load parameters are identified through a linear parametric

(ARMAX) model using a Kalman Filter.

But identifying non-linear dynamics through a linear model does not guarantee an accurate

functional representation.

The ability of a multi-layer perceptron type neural network to learn a large class of non-linear

functions is well known. The neural network identifier evolves through the learning of a

suitable time sequence of input/output patterns generated by the motor model. The ability to

successfully train without explicit knowledge of the motor/load dynamics is the key

advantage in this type of identification. Moreover the superior generalizing capability of the

neural network allows accurate emulation of the motor dynamics for previously untrained

inputs. The inherent noise rejection property of the neural network is useful in dealing with

noisy input/output characteristics.

22
 Chapter 4

OBJECTIVE

23
The neural networks role is to identify the unknown mapping between the voltage command

and the speed of the motor. The trained neural network is then used for controller design.

4.1 SYSTEM DYNAMICS:

K(t) = RaIa(t) La [dia(t)/dt] + Vt(t) (1)

Kia(t) = J[d(t)/dt] + D(t) + TL(t) (2)

Where,

(t) - rotor speed (rad/s)

Vt(t) - terminal voltage (V)

Ia(t) - armature current (A)

TL(t) - load torque (Nm)

J - Rotor inertia (Nm2)

K - Torque & back emf constant (Nm/A)

D - viscose friction coefficient (Nms)

Ra - armature resistance ()

La - armature inductance (H)

4.2 CONTROL SYSTEM APPLICATIONS:

Predictive control method

NARMA-L2 control method

Model reference control method.

All are based on standard linear control architecture.

Two steps involved when used for control:

System identification

24
 Control design

SI stage, a neural network model of the plant that we want to control developed.

Control design stage, the neural network plant model used to train the controller.

4.3 SYSTEM IDENTIFICATION:

It is to train a neural network to represent the forward dynamics of the plant. The prediction

error between the plant output and the neural network output is used as the neural network

training signal.

Figure 14: Plant Identification

The standard model used for non-linear identification in NARMA model.

Figure 15: Neural Network Plant Model

TDL- Tapped Delay Lines that stores previous values of the input signal

IW (i, j) - Weight matrix from layer number i to j

IW (j, i) - Weight matrix from layer number j to i

25
4.4 CONTROL DESIGN:

We use Model reference Adaptive Control (MRAC) strategy. This architecture consists of

two neural networks;

Controller network

Plant model network

After SI, the controller is trained so that the plant output follows the reference model output.

The controller training is done separately and requires the use of dynamic back-propagation.

DBP gets applied to a larger class of plant and hence the controller requires minimal online

computation. Generally indirect MRAC is preferred.

Figure 16: Model Reference Control Architecture

4.5 DISCRETE TIME DC MOTOR MODEL:

TL(t) = WP2(t) [sign(Wp(t)] ;

where is a constant.

The discrete time model speed equation governing the system dynamics is given by;

26
 4.6 IDENTIFICATION OF DC MOTOR MODEL:

The equation can be manipulated to the form;

Vt(k) = g[Wp(k+1), Wp(k), Wp(k-1)]

where the function g[.] is given by;

g[Wp(k+1), Wp(k), Wp(k-1)] = { Wp(k+1) - Wp(k) - Wp(k-1) - [sign (Wp(k))]

Wp2(k) - [sign (Wp(k-1))] Wp2(k-1) }/ ;

and is assumed to be unknown. An ANN is trained to emulate the unknown function g[.].

However as Wp(k+1) cannot be readily available.

Figure 17: Structure Of The ANN Identifier

4.7 TRAJECTORY CONTROL OF DC MOTOR USING TRAINED ANN:

The objective of the control system is to drive the motor so that its speed Wp(k), follows a

pre-specified trajectory, Wm(k). The following second order reference model is chosen;

Wm(k+1) = 0.6 Wm(k) + 0.2 Wm(k-1) + r(k) ;

27
For a given desired sequence {Wm(k)}, the corresponding control sequence {r(k)} can be

calculated using the above relation.

The speed at (k+1)th time step can be predicted from;

= 0.6 Wp(k) + 0.2 Wp(k-1) + r(k) ;

This result can be fed to the ANN to estimate the control input at kth time step using;

= N [Wp(k), Wp(k-1), Wp(k-2)] ;

28
 Chapter 5

PROBLEM STATEMENT

29
A separately excited dc motor with name plate ratings of 1 hp, 220V, 550 rpm is used in all

simulations. Following parameter values are associated with it.

J = 0.068 Kgm2

K = 3.475 Nm A-1

Ra = 7.56

La= 0.055 H

D = 0.03475 Nms

= 0.0039 Nms2

T = 40ms

As from the system dynamics equations, are constant values based on the motor

parameters J, K, D, Ra, La and the sampling period T. While in addition to being

functions of the above parameters are also functions of . The value k denotes the kth time

step.

= 2*(La J + Ra J + La D Ra D T K2 T)/( La J + 2*Ra J + 2*La D) ;

= (-) La J/( La J + 2*Ra J + 2*La D) ;

= (-) 2* (La + Ra T)/( La J + 2*Ra J + 2*La D) ;

= 2*La /( La J + 2*Ra J + 2*La D) ;

= 2*K*T/( La J + 2*Ra J + 2*La D) ;

Using the given values of the system parameters in the above stated equations we get;

= 0.0506 ;

= (-)0.003611 ;

30
 = (-)0.002692 ;

= 0.000414 ;

= 0.26795 ;

The system equation becomes;

(K2 + Ra D) W(t) = K Vt(t) J La (Ra J + La D) La Ra TL(t) ;

LIMITS TAKEN INTO CONSIDERATION :

-30.0 < WP(k) < 30.0 rad/s

-1.0 < Wp(k- 1) - Wp(k) < 1.0 rad/s

-100 v < Vt(k) < 100 v

31
 Chapter 6

GENERATION OF THE SIGNAL

32
6.1 OVERVIEW:

= N [ Wp(k), Wp(k-1) ] + Vt(k) ;

So, now the system dynamic equations get converted into two first order differential

equations as follows :

(t) = (Vt(t) / La) (Ra ia(t) / La) (K W(t) / La) ;

(t) = (K ia(t) / J) (D W(t) / J) (TL(t) / J) ;

TAKING INITIAL VALUES AS,

Wp(0) = W(0) = 0;

ia(0) = 0;

Substituting the values of the motor parameters in the above two equations along with the

initial conditions, the above two equations get transformed into:

= 18.182 Vt 137.454 y1 63.182 y2 ;

= 51.1 y1 0.511 y2 14.7 TL ;

Considering y1 = ia(t) and y2 = W(t)

Vt(k) = 50 + 45 is assumed.

33
6.2 RESULT :

Figure 18: Actual And Estimated Rotor Speeds

34
 Chapter 7

SYSTEM IDENTIFICATION

35
7.1 OVERVIEW:

The 3-layer feed-forward ANN used :

Figure 19: Three-Layer Feed forward ANN

As discussed earlier, the motor dynamics can be partitioned as;

Wp(k+1) = f[ Wp(k), Wp(k-1) ] + Vt(k) ;

Where the function f[.] is given by;

f[ Wp(k), Wp(k-1) ] =

which is supposed to be an UNKNOWN NON-LINEAR FUNCTION.

The values Wp(k) and Wp(k - 1) which are the independent variables of f [.], are selected as

the inputs to the ANN. The corresponding ANN output target f [Wp(k), Wp(k-1)] is given by

above equation.

36
The ANN is trained off-line using randomly generated input patterns of [Wp(k), Wp(k-1)] and

the corresponding target f [Wp(k), Wp(k-1)]. The choice of Wp(k), Wp(k-1) has to satisfy the

constraints as specified earlier.

The motor speed is estimated by the trained ANN predictor as

= N [ Wp(k), Wp(k-1) ] + Vt(k) ;

Where N[.] denotes the ANN output for a given set of . Inputs.

ANN topology and the training effort are briefly described by the following statistics:

Number of inputs 2

Number of outputs 1

Number of Hidden layers 1

Number of Hidden neurons 5

Number of training patterns P 500

Number of training sweeps 1000

Learning Step 0.1

Momentum Gain 0.1

Etotal threshold 0.04

As mentioned earlier, except for the number of inputs/outputs of the ANN, all other design

and learning parameters are selected by trial and error.

37
7.2 BLOCK DIAGRAM :

Figure 20: Structure Of The ANN For Identification Of The Dc Motor

The trained ANN is applied as a series-parallel type identifier to estimate the value of the

function f [.]. z-1 in any figure indicates a unit time delay. The performance of the trained

ANN identifier is evaluated by comparing the actual and estimated speeds as calculated from

the afore mentioned equations above for the following arbitrarily selected terminal voltage

sequence,

Vt(k) = 50 + 45 .

7.3 C-CODE FOR TRAINING WITH BACK-PROPAGATION OF THE ANN:

A single neuron (i.e. processing unit) takes it total input In and produces an output activation

Out. I shall take this to be the sigmoid function

Out = 1.0/(1.0 + exp(-In)); /* Out = Sigmoid(In) */

though other activation functions are often used (e.g. linear or hyperbolic tangent). This has

the effect of squashing the infinite range of In into the range 0 to 1. It also has the convenient

property that its derivative takes the particularly simple form

Sigmoid_Derivative = Sigmoid * (1.0 - Sigmoid) ;

38
Typically, the input In into a given neuron will be the weighted sum of output activations

feeding in from a number of other neurons. It is convenient to think of the activations flowing

through layers of neurons. So, if there are NumUnits1 neurons in layer 1, the total activation

flowing into our layer 2 neuron is just the sum over Layer1Out[i]*Weight[i], where

Weight[i] is the strength/weight of the connection between unit i in layer 1 and our unit in

layer 2. Each neuron will also have a bias, or resting state, that is added to the sum of inputs,

and it is convenient to call this weight[0]. We can then write

Layer2In=Weight[0]; /* start with the bias */

for( i = 1 ; i <= NumUnits1 ; i++ ) { /* i loop over layer 1 units */

Layer2In += Layer1Out[i] * Weight[i] ; /* add in weighted contributions

from layer 1 */

Layer2Out = 1.0/(1.0 + exp(-Layer2In)) ; /* compute sigmoid to give activation */

Normally layer 2 will have many units as well, so it is appropriate to write the weights

between unit i in layer 1 and unit j in layer 2 as an array Weight[i][j]. Thus to get the output

of unit j in layer 2 we have

Layer2In[j]=Weight[0][j];

for( i = 1 ; i <= NumUnits1 ; i++ ) {

Layer2In[j] += Layer1Out[i] * Weight[i][j] ;

Layer2Out[j] = 1.0/(1.0 + exp(-Layer2In[j])) ;

Remember that in C the array indices start from zero, not one, so we would declare our

variables as

39
 double Layer1Out[NumUnits1+1] ;

double Layer2In[NumUnits2+1] ;

double Layer2Out[NumUnits2+1] ;

double Weight[NumUnits1+1][NumUnits2+1] ;

(or, more likely, declare pointers and use calloc or malloc to allocate the memory). Naturally,

we need another loop to get all the layer 2 outputs

for( j = 1 ; j <= NumUnits2 ; j++ ) {

Layer2In[j] = Weight[0][j] ;

for( i = 1 ; i <= NumUnits1 ; i++ ) {

Layer2In[j] += Layer1Out[i] * Weight[i][j] ;

Layer2Out[j] = 1.0/(1.0 + exp(-Layer2In[j])) ;

Three layer networks are necessary and sufficient for most purposes, so our layer 2 outputs

feed into a third layer in the same way as above

for( j = 1 ; j <= NumUnits2 ; j++ ) { /* j loop computes layer 2 activations */

Layer2In[j] = Weight12[0][j] ;

for( i = 1 ; i <= NumUnits1 ; i++ ) {

Layer2In[j] += Layer1Out[i] * Weight12[i][j] ;

Layer2Out[j] = 1.0/(1.0 + exp(-Layer2In[j])) ;

for( k = 1 ; k <= NumUnits3 ; k++ ) { /* k loop computes layer 3 activations */

40
 Layer3In[k] = Weight23[0][k] ;

for( j = 1 ; j <= NumUnits2 ; j++ ) {

Layer3In[k] += Layer2Out[j] * Weight23[j][k] ;

Layer3Out[k] = 1.0/(1.0 + exp(-Layer3In[k])) ;

The code can start to become confusing at this point - I find that keeping a separate index i, j,

k for each layer helps, as does an intuitive notation for distinguishing between the different

layers of weights Weight12 and Weight23. For obvious reasons, for three layer networks, it is

traditional to call layer 1 the Input layer, layer 2 the Hidden layer, and layer 3 the Output

layer.

Also, to save getting all the In's and Out's confused, we can write LayerNIn as SumN. Our

code can thus be written

for( j = 1 ; j <= NumHidden ; j++ ) { /* j loop computes hidden unit activations

*/

SumH[j] = WeightIH[0][j] ;

for( i = 1 ; i <= NumInput ; i++ ) {

SumH[j] += Input[i] * WeightIH[i][j] ;

Hidden[j] = 1.0/(1.0 + exp(-SumH[j])) ;

41
 }

for( k = 1 ; k <= NumOutput ; k++ ) { /* k loop computes output unit activations

*/

SumO[k] = WeightHO[0][k] ;

for( j = 1 ; j <= NumHidden ; j++ ) {

SumO[k] += Hidden[j] * WeightHO[j][k] ;

Output[k] = 1.0/(1.0 + exp(-SumO[k])) ;

Generally we will have a whole set of NumPattern training patterns, i.e. pairs of input and

target output vectors,

Input[p][i] , Target[p][k]

labelled by the index p. The network learns by minimizing some measure of the error of the

network's actual outputs compared with the target outputs. For example, the sum squared

error overall output units k and all training patterns p will be given by

Error = 0.0 ;

for( p = 1 ; p <= NumPattern ; p++ ) {

for( k = 1 ; k <= NumOutput ; k++ ) {

Error += 0.5 * (Target[p][k] - Output[p][k]) * (Target[p][k] -

Output[p][k]) ;

42
(The factor of 0.5 is conventionally included to simplify the algebra in deriving the learning

algorithm.) If we insert the above code for computing the network outputs into the p loop of

this, we end up with

Error = 0.0 ;

for( p = 1 ; p <= NumPattern ; p++ ) { /* p loop over training patterns */

for( j = 1 ; j <= NumHidden ; j++ ) { /* j loop over hidden units */

SumH[p][j] = WeightIH[0][j] ;

for( i = 1 ; i <= NumInput ; i++ ) {

SumH[p][j] += Input[p][i] * WeightIH[i][j] ;

Hidden[p][j] = 1.0/(1.0 + exp(-SumH[p][j])) ;

for( k = 1 ; k <= NumOutput ; k++ ) { /* k loop over output units */

SumO[p][k] = WeightHO[0][k] ;

for( j = 1 ; j <= NumHidden ; j++ ) {

SumO[p][k] += Hidden[p][j] * WeightHO[j][k] ;

Output[p][k] = 1.0/(1.0 + exp(-SumO[p][k])) ;

Error += 0.5 * (Target[p][k] - Output[p][k]) * (Target[p][k] -

Output[p][k]) ; /* Sum Squared Error */

The next stage is to iteratively adjust the weights to minimize the network's error. A standard

way to do this is by 'gradient descent' on the error function. We can compute how much the

43
error is changed by a small change in each weight (i.e. compute the partial derivatives

dError/dWeight) and shift the weights by a small amount in the direction that reduces the

error.

Error = 0.0 ;

for( p = 1 ; p <= NumPattern ; p++ ) { /* repeat for all the training patterns */

for( j = 1 ; j <= NumHidden ; j++ ) { /* compute hidden unit activations */

SumH[p][j] = WeightIH[0][j] ;

for( i = 1 ; i <= NumInput ; i++ ) {

SumH[p][j] += Input[p][i] * WeightIH[i][j] ;

Hidden[p][j] = 1.0/(1.0 + exp(-SumH[p][j])) ;

for( k = 1 ; k <= NumOutput ; k++ ) { /* compute output unit activations and

errors */

SumO[p][k] = WeightHO[0][k] ;

for( j = 1 ; j <= NumHidden ; j++ ) {

SumO[p][k] += Hidden[p][j] * WeightHO[j][k] ;

Output[p][k] = 1.0/(1.0 + exp(-SumO[p][k])) ;

Error += 0.5 * (Target[p][k] - Output[p][k]) * (Target[p][k] - Output[p][k]) ;

DeltaO[k] = (Target[p][k] - Output[p][k]) * Output[p][k] * (1.0 -

Output[p][k]) ;

for( j = 1 ; j <= NumHidden ; j++ ) { /* 'back-propagate' errors to hidden layer */

44
 SumDOW[j] = 0.0 ;

for( k = 1 ; k <= NumOutput ; k++ ) {

SumDOW[j] += WeightHO[j][k] * DeltaO[k] ;

DeltaH[j] = SumDOW[j] * Hidden[p][j] * (1.0 - Hidden[p][j]) ;

for( j = 1 ; j <= NumHidden ; j++ ) { /* update weights WeightIH */

DeltaWeightIH[0][j] = eta * DeltaH[j] + alpha * DeltaWeightIH[0][j] ;

WeightIH[0][j] += DeltaWeightIH[0][j] ;

for( i = 1 ; i <= NumInput ; i++ ) {

DeltaWeightIH[i][j] = eta * Input[p][i] * DeltaH[j] + alpha *

DeltaWeightIH[i][j];

WeightIH[i][j] += DeltaWeightIH[i][j] ;

for( k = 1 ; k <= NumOutput ; k ++ ) { /* update weights WeightHO */

DeltaWeightHO[0][k] = eta * DeltaO[k] + alpha * DeltaWeightHO[0][k] ;

WeightHO[0][k] += DeltaWeightHO[0][k] ;

for( j = 1 ; j <= NumHidden ; j++ ) {

DeltaWeightHO[j][k] = eta * Hidden[p][j] * DeltaO[k] + alpha *

DeltaWeightHO[j][k] ;

WeightHO[j][k] += DeltaWeightHO[j][k] ;

45
he weight changes DeltaWeightIH and DeltaWeightHO are each made up of two components.

First, the eta component that is the gradient descent contribution. Second, the alpha

component that is a 'momentum' term which effectively keeps a moving average of the

gradient descent weight change contributions, and thus smoothes out the overall weight

changes. Fixing good values of the learning parameters eta and alpha is usually a matter of

trial and error. Certainly alpha must be in the range 0 to 1, and a non-zero value does usually

speed up learning. Finding a good value for eta will depend on the problem, and also on the

value chosen for alpha. If it is set too low, the training will be unnecessarily slow. Having it

too large will cause the weight changes to oscillate wildly, and can slow down or even

prevent learning altogether.

The complete training process will consist of repeating the above weight updates for a

number of epochs (using another for loop) until some error crierion is met, for example the

Error falls below some chosen small number.

for( epoch = 1 ; epoch < LARGENUMBER ; epoch++ ) {

/* ABOVE CODE FOR ONE ITERATION */

if( Error < SMALLNUMBER ) break ;

If the training patterns are presented in the same systematic order during each epoch, it is

possible for weight oscillations to occur. It is therefore generally a good idea to use a new

random order for the training patterns for each epoch. If we put the NumPattern training

pattern indices p in random order into an array ranpat[], then it is simply a matter of

replacing our training pattern loop

46
 for( p = 1 ; p <= NumPattern ; p++ ) {

with

for( np = 1 ; np <= NumPattern ; np++ ) {

p = ranpat[np] ;

Generating the random array ranpat[] is not quite so simple, but the following code will do

the job

for( p = 1 ; p <= NumPattern ; p++ ) { /* set up ordered array */

ranpat[p] = p ;

for( p = 1 ; p <= NumPattern ; p++) { /* swap random elements into each

position */

np = p + rando() * ( NumPattern + 1 - p ) ;

op = ranpat[p] ; ranpat[p] = ranpat[np] ; ranpat[np] = op ;

Naturally, one must set some initial network weights to start the learning process. Starting all

the weights at zero is generally not a good idea, as that is often a local minimum of the error

function. It is normal to initialize all the weights with small random values. If rando() is your

favourite random number generator function that returns a flat distribution of random

numbers in the range 0 to 1, and smallwt is the maximum absolute size of your initial

weights, then an appropriate section of weight initialization code would be

for( j = 1 ; j <= NumHidden ; j++ ) { /* initialize WeightIH and DeltaWeightIH

*/

47
 for( i = 0 ; i <= NumInput ; i++ ) {

DeltaWeightIH[i][j] = 0.0 ;

WeightIH[i][j] = 2.0 * ( rando() - 0.5 ) * smallwt ;

for( k = 1 ; k <= NumOutput ; k ++ ) { /* initialize WeightHO and

DeltaWeightHO */

for( j = 0 ; j <= NumHidden ; j++ ) {

DeltaWeightHO[j][k] = 0.0 ;

WeightHO[j][k] = 2.0 * ( rando() - 0.5 ) * smallwt ;

48
 Chapter 8

TRAJECTORY CONTROL

49
8.1 OVERVIEW:

The objective of the control system is to drive the motor so that its speed, Wp(k), follows a

prespecified trajectory,Wm(k). This is done by letting the dc motor follow the output of a

selected reference model throughout the trajectory. The following second order reference

model is selected;

Wm(k+ 1) = 0.6 Wm(k) + 0.2 Wm(k - 1) + r(k) ;

r(k) is the bounded input to the reference model. The coefficients are selected to ensure that

its poles are within the unit circle and has the type of response that can be achieved by the dc

motor. For a given desired sequence (Wm(k)) (trajectory), the corresponding

control sequence (r(k)) can be calculated using the above equation.

The controller topology trained in the previous chapter is now used to estimate the motor

terminal voltage Vt(k) which enables accurate trajectory control of the shaft speed Wp(k).

Performance of the controller is simulated for arbitrarily selected speed tracks (Wm ( k ) ). A

graphical comparison of the specified and actual speed trajectories are hence presented.

Let's for a moment assume that the tracking error Ec(k) is zero, and that the nonlinear

function f[.] is known. The control input Vt(k) to the motor at the kth time step can be

calculated as

Vt(k) = [-f[Wp(k), Wp(k-1)] + 0.6*Wp(k) + 0.2*Wp(k-1) + r(k)]/ ;

Hence the tracking error difference equation becomes

Ec(k+1) = 0.6*Ec(k) + 0.2*Ec(k-1) ;

50
Since the reference model is asymptotically stable, it follows that (k+l) = 0 for

arbitrary initial conditions. However since f[.] is not known, its estimated value from the

ANN trained under the controller topology is used to estimate the controlled terminal voltage.

(k) = [-N[Wp(k), Wp(k-1)] + 0.6*Wp(k) + 0.2*Wp(k-1) + r(k)]/ ;

The tracking control capability of the model was investigated for different arbitrarily

specified trajectories. Only a specific result is shown for brevity. In this case, the

specified speed trajectory is defined by

Wm( k) = 10*Sin(2.0* kT/4) + 16*Sin(2.0* kT/7) [0,20] ;

For the above trajectory, the corresponding (r(k)) is derived by using the first equation. This

is applied to the model shown below. The matrix corresponds to the reference model

coefficients [0.6 0.2].

51
8.3 PLOTS:

Figure 21: Actual And Estimated Terminal Voltage Of The DC Motor

Figure 22: Tracking Performance For A Sinusoidal Reference Track

52
8.4 OVERALL CONTROLLER STRUCTURE :

Figure 23: The Overall Structure Of The Controller In Topology

The maximum tracking error is 0.55 rad/s.

Both the ANN is the Block Diagram can be trained using the C-code mentioned in Section

7.4.

The overall structure of the identification and control system is displayed in Figure 20.

53
 Chapter 9

RESULTS

54
 9.1 TABLE OF OUTPUTS FOR COMPARISON:

KT Wp(k) (k+1) RMS ERROR Wm(k) Vt(k) (k) Wp(k)

0 0 0 0 0 0 0 0
0.04 0.4478 0 0.4478 1.2022 5.5603 0 0
0.08 1.6429 1.5126 0.1303 2.4013 11.0919 7.1898 1.9486
0.12 3.0828 3.0041 0.0787 3.5939 16.5662 10.8739 2.988
0.16 4.5727 4.4738 0.0989 4.7769 21.9551 15.6482 4.3185
0.2 6.0509 5.919 0.1319 5.9471 27.231 20.7295 5.7224
0.24 7.4945 7.3335 0.161 7.1014 32.367 25.8374 7.1229
0.28 8.8921 8.7104 0.1817 8.2368 37.3371 30.8718 8.4934
0.32 10.2362 10.0428 0.1934 9.3503 42.1166 35.7857 9.8221
0.36 11.5206 11.324 0.1966 10.4389 46.6817 40.5486 11.1015
0.4 12.7401 12.5478 0.1923 11.4998 51.0103 45.1353 12.3261
0.44 13.8896 13.7082 0.1814 12.5303 55.0816 49.5232 13.4908
0.48 14.9648 14.7996 0.1652 13.5277 58.8768 53.6916 14.5913
0.52 15.9619 15.8168 0.1451 14.4895 62.3785 57.621 15.6236
0.56 16.8774 16.7554 0.122 15.4132 65.5715 61.2935 16.5841
0.6 17.7082 17.611 0.0972 16.2966 68.4425 64.6931 17.4695
0.64 18.4521 18.3802 0.0719 17.1374 70.9802 67.8052 18.2773
0.68 19.1069 19.06 0.0469 17.9338 73.1756 70.6172 19.0049
0.72 19.6712 19.6478 0.0234 18.6837 75.0219 73.1186 19.6505
0.76 20.144 20.1419 0.0021 19.3855 76.5142 75.3006 20.2126
0.8 20.5248 20.5409 -0.0161 20.0376 77.6504 77.1566 20.6902
0.84 20.8136 20.8443 -0.0307 20.6386 78.4303 78.6822 21.0826
0.88 21.0108 21.052 -0.0412 21.1872 78.856 79.8748 21.3897
0.92 21.1172 21.1646 -0.0474 21.6825 78.9319 80.734 21.6116
0.96 21.1342 21.1833 -0.0491 22.1234 78.6646 81.2614 21.7489
1 21.0634 21.1098 -0.0464 22.5093 78.0627 81.4605 21.8027
1.04 20.9073 20.9465 -0.0392 22.8396 77.1371 81.3367 21.7742
1.08 20.6684 20.6962 -0.0278 23.1139 75.9006 80.8977 21.6654
1.12 20.3501 20.3625 -0.0124 23.3321 74.3677 80.1531 21.4783
1.16 19.9556 19.9493 0.0063 23.4941 72.5548 79.1143 21.2158
1.2 19.4888 19.4609 0.0279 23.6001 70.4801 77.7939 20.8806
1.24 18.954 18.9023 0.0517 23.6504 68.1631 76.2061 20.4759
1.28 18.3559 18.2787 0.0772 23.6455 65.6248 74.3667 20.0054
1.32 17.6994 17.5958 0.1036 23.586 62.8873 72.2931 19.473
1.36 16.9896 16.8595 0.1301 23.4729 59.974 70.0038 18.883
1.4 16.2323 16.0761 0.1562 23.3071 56.9089 67.5177 18.2398
1.44 15.4331 15.2521 0.181 23.0897 53.7168 64.8554 17.5482
1.48 14.5982 14.3942 0.204 22.8221 50.4232 62.0378 16.8132
1.52 13.7338 13.5093 0.2245 22.5057 47.0539 59.0865 16.0401
1.56 12.8462 12.6042 0.242 22.1421 43.6346 56.0235 15.2342
1.6 11.9421 11.6859 0.2562 21.733 40.1915 52.8713 14.4011
1.64 11.028 10.7614 0.2666 21.2803 36.75 49.6521 13.5466
1.68 10.1105 9.8377 0.2728 20.786 33.3357 46.3886 12.6764
1.72 9.1966 8.9214 0.2752 20.252 29.9733 43.1025 11.7963
1.76 8.2928 8.0193 0.2735 19.6806 26.6868 39.8165 10.9123
1.8 7.4055 7.1377 0.2678 19.0741 23.4993 36.5515 10.0303

55
1.84 6.5413 6.283 0.2583 18.4347 20.433 33.3286 9.156
1.88 5.7063 5.461 0.2453 17.765 17.5087 30.1678 8.2952
1.92 4.9067 4.6773 0.2294 17.0674 14.7459 27.0884 7.4535
1.96 4.1481 3.937 0.2111 16.3445 12.1626 24.1089 6.6363
2 3.4359 3.2451 0.1908 15.5988 9.7753 21.2467 5.8489
2.04 2.775 2.6059 0.1691 14.8331 7.5984 18.5178 5.0961
2.08 2.1701 2.0234 0.1467 14.0499 5.6448 15.9369 4.3825
2.12 1.6252 1.5008 0.1244 13.2519 3.9255 13.5179 3.7124
2.16 1.1436 1.0411 0.1025 12.4418 2.4492 11.2729 3.0898
2.2 0.7287 0.6468 0.0819 11.6223 1.2229 9.2124 2.518
2.24 0.3824 0.3195 0.0629 10.796 0.2512 7.3459 2.0002
2.28 0.1068 0.0605 0.0463 9.9656 -0.4632 5.6808 1.5387
2.32 -0.0972 -0.1294 0.0322 9.1336 -0.9199 4.2232 1.1357
2.36 -0.229 -0.2503 0.0213 8.3027 -1.1205 2.9773 0.7925
2.4 -0.2889 -0.3024 0.0135 7.4753 -1.0689 1.9454 0.5102
2.44 -0.278 -0.287 0.009 6.6539 -0.7708 1.1295 0.2891
2.48 -0.1978 -0.2055 0.0077 5.8409 -0.2343 0.5291 0.1289
2.52 -0.0507 -0.0601 0.0094 5.0385 0.5309 0.1412 0.0289
2.56 0.1608 0.1465 0.0143 4.249 1.5129 -0.0387 -0.0122
2.6 0.4332 0.4111 0.0221 3.4745 2.6982 -0.0167 0.0038
2.64 0.7626 0.7302 0.0324 2.7171 4.0715 0.2008 0.0746
2.68 1.1443 1.0996 0.0447 1.9786 5.6159 0.6048 0.1976
2.72 1.5735 1.5147 0.0588 1.2608 7.3131 1.1848 0.3693
2.76 2.0448 1.9707 0.0741 0.5654 9.1434 1.9288 0.5862
2.8 2.5523 2.4624 0.0899 -0.106 11.0861 2.8237 0.8442
2.84 3.0902 2.984 0.1062 -0.752 13.1193 3.8549 1.1389
2.88 3.652 3.5298 0.1222 -1.3712 15.2203 5.0068 1.4657
2.92 4.2313 4.0937 0.1376 -1.9626 17.3657 6.2626 1.8197
2.96 4.8216 4.6693 0.1523 -2.5251 19.5318 7.6046 2.1959
3 5.4161 5.2504 0.1657 -3.0579 21.6942 9.0146 2.589
3.04 6.0083 5.8304 0.1779 -3.5601 23.8286 10.4735 2.9937
3.08 6.5911 6.4028 0.1883 -4.0312 25.9108 11.962 3.4047
3.12 7.1583 6.9611 0.1972 -4.4707 27.9166 13.4597 3.8164
3.16 7.7033 7.4988 0.2045 -4.8783 29.8224 14.947 4.2235
3.2 8.2199 8.0095 0.2104 -5.254 31.605 16.4038 4.6208
3.24 8.7018 8.4871 0.2147 -5.5976 33.2421 17.8104 5.0031
3.28 9.1434 8.9256 0.2178 -5.9092 34.7122 19.147 5.3652
3.32 9.5392 9.3191 0.2201 -6.1892 35.9949 20.3946 5.7023
3.36 9.8837 9.6623 0.2214 -6.4379 37.0713 21.535 6.0096
3.4 10.1723 9.95 0.2223 -6.6559 37.9235 22.5504 6.2827
3.44 10.4001 10.1774 0.2227 -6.8439 38.5353 23.4241 6.5174
3.48 10.5634 10.3403 0.2231 -7.0025 38.8923 24.1403 6.7096
3.52 10.6584 10.4347 0.2237 -7.1327 38.9816 24.6849 6.8557
3.56 10.6815 10.4572 0.2243 -7.2355 38.7923 25.0448 6.9523
3.6 10.6303 10.4049 0.2254 -7.3121 38.3156 25.208 6.9964
3.64 10.5021 10.2753 0.2268 -7.3636 37.5444 25.1647 6.9852
3.68 10.2951 10.0667 0.2284 -7.3914 36.474 24.9062 6.9165
3.72 10.0078 9.7778 0.23 -7.3968 35.1017 24.4256 6.7882
3.76 9.6394 9.4078 0.2316 -7.3815 33.427 23.7178 6.5987
3.8 9.1892 8.9566 0.2326 -7.3468 31.4516 22.7794 6.3467

56
3.84 8.6575 8.4248 0.2327 -7.2944 29.1795 21.6088 6.0314
3.88 8.0445 7.8133 0.2312 -7.226 26.6167 20.2061 5.6523
3.92 7.3515 7.124 0.2275 -7.1433 23.7715 18.5735 5.2094
3.96 6.58 6.3589 0.2211 -7.0481 20.6541 16.7146 4.7029
4 5.7319 5.521 0.2109 -6.9421 17.277 14.635 4.1337
4.04 4.8097 4.6135 0.1962 -6.8272 13.6544 12.342 3.5028
4.08 3.8166 3.6403 0.1763 -6.7052 9.8028 9.8442 2.8118
4.12 2.756 2.6059 0.1501 -6.5778 5.7399 7.1525 2.0626
4.16 1.6319 1.515 0.1169 -6.447 1.4856 4.2786 1.2577
4.2 0.4488 0.373 0.0758 -6.3144 -2.9389 1.2357 0.3997
4.24 -0.7879 -0.8145 0.0266 -6.1819 -7.5111 -1.9615 -0.5082
4.28 -2.0663 -2.0412 -0.0251 -6.0512 -12.2071 -5.3086 -1.4622
4.32 -3.3742 -3.301 -0.0732 -5.9241 -17.0019 -8.8219 -2.4543
4.36 -4.7023 -4.5872 -0.1151 -5.8021 -21.8697 -12.4678 -3.4751
4.4 -6.0428 -5.893 -0.1498 -5.6869 -26.7837 -16.2198 -4.517
4.44 -7.3881 -7.2112 -0.1769 -5.5799 -31.7168 -20.0556 -5.5736
4.48 -8.73 -8.5344 -0.1956 -5.4827 -36.6413 -23.9547 -6.6389
4.52 -10.0622 -9.8555 -0.2067 -5.3967 -41.5293 -27.894 -7.7067
4.56 -11.376 -11.1667 -0.2093 -5.323 -46.353 -31.8537 -8.7718
4.6 -12.6646 -12.4607 -0.2039 -5.263 -51.0847 -35.8092 -9.8279
4.64 -13.9215 -13.73 -0.1915 -5.2177 -55.697 -39.7376 -10.8693
4.68 -15.1394 -14.9672 -0.1722 -5.1881 -60.163 -43.6175 -11.8907
4.72 -16.3117 -16.1652 -0.1465 -5.1751 -64.4566 -47.4253 -12.8866
4.76 -17.4326 -17.3169 -0.1157 -5.1795 -68.5526 -51.1379 -13.8517
4.8 -18.4963 -18.4156 -0.0807 -5.2019 -72.4267 -54.7335 -14.7811
4.84 -19.4972 -19.4547 -0.0425 -5.2429 -76.0561 -58.1911 -15.6701
4.88 -20.4302 -20.428 -0.0022 -5.3028 -79.4191 -61.4899 -16.5143
4.92 -21.2907 -21.3298 0.0391 -5.3821 -82.4956 -64.61 -17.3094
4.96 -22.0744 -22.1547 0.0803 -5.4807 -85.2673 -67.5326 -18.0514
5 -22.7773 -22.8976 0.1203 -5.5988 -87.7175 -70.2399 -18.7368
5.04 -23.396 -23.5542 0.1582 -5.7363 -89.8316 -72.7158 -19.3621
5.08 -23.9276 -24.1205 0.1929 -5.8929 -91.5967 -74.9454 -19.9245
5.12 -24.3696 -24.5931 0.2235 -6.0683 -93.0023 -76.9154 -20.4211
5.16 -24.7197 -24.969 0.2493 -6.262 -94.0399 -78.6142 -20.8497
5.2 -24.9765 -25.2461 0.2696 -6.4733 -94.7031 -80.0318 -21.2082
5.24 -25.1388 -25.4225 0.2837 -6.7016 -94.9881 -81.1602 -21.4951
5.28 -25.2058 -25.4974 0.2916 -6.9459 -94.8931 -81.993 -21.7089
5.32 -25.1774 -25.4701 0.2927 -7.2054 -94.4186 -82.5257 -21.8487
5.36 -25.0537 -25.3409 0.2872 -7.4788 -93.5675 -82.756 -21.914
5.4 -24.8356 -25.1105 0.2749 -7.765 -92.345 -82.683 -21.9044
5.44 -24.5241 -24.7803 0.2562 -8.0628 -90.7585 -82.3081 -21.8201
5.48 -24.1207 -24.3523 0.2316 -8.3706 -88.8174 -81.6344 -21.6615
5.52 -23.6276 -23.8291 0.2015 -8.687 -86.5336 -80.6666 -21.4293
5.56 -23.047 -23.2138 0.1668 -9.0103 -83.9207 -79.4113 -21.1247
5.6 -22.3821 -22.5102 0.1281 -9.3391 -80.9943 -77.8768 -20.7492
5.64 -21.6363 -21.7224 0.0861 -9.6713 -77.7722 -76.0734 -20.3045
5.68 -20.8131 -20.8553 0.0422 -10.0054 -74.2734 -74.0131 -19.793
5.72 -19.9167 -19.9139 -0.0028 -10.3394 -70.5188 -71.7085 -19.217
5.76 -18.9517 -18.904 -0.0477 -10.6714 -66.5308 -69.1739 -18.5793
5.8 -17.923 -17.8315 -0.0915 -10.9994 -62.3329 -66.4253 -17.883

57
5.84 -16.8358 -16.7028 -0.133 -11.3214 -57.9499 -63.4792 -17.1317
5.88 -15.6958 -15.5246 -0.1712 -11.6354 -53.4074 -60.3535 -16.3288
5.92 -14.5088 -14.3039 -0.2049 -11.9395 -48.7321 -57.0667 -15.4786
5.96 -13.2809 -13.0477 -0.2332 -12.2314 -43.9509 -53.638 -14.5851
6 -12.0187 -11.7635 -0.2552 -12.5093 -39.0916 -50.0874 -13.6529
6.04 -10.7289 -10.4585 -0.2704 -12.7711 -34.1818 -46.435 -12.6866
6.08 -9.4183 -9.1404 -0.2779 -13.0147 -29.2496 -42.7014 -11.6913
6.12 -8.0939 -7.8165 -0.2774 -13.2382 -24.3226 -38.9075 -10.6719
6.16 -6.7628 -6.4944 -0.2684 -13.4397 -19.4284 -35.0734 -9.6338
6.2 -5.4326 -5.1813 -0.2513 -13.6172 -14.5938 -31.2195 -8.5822
6.24 -4.1103 -3.8846 -0.2257 -13.769 -9.8452 -27.3667 -7.5229
6.28 -2.8031 -2.6113 -0.1918 -13.8932 -5.208 -23.5341 -6.4611
6.32 -1.5184 -1.3682 -0.1502 -13.9882 -0.7067 -19.7409 -5.4024
6.36 -0.2634 -0.1618 -0.1016 -14.0524 3.6354 -16.0057 -4.3525
6.4 0.954 1.0016 -0.0476 -14.0842 7.7966 -12.3466 -3.3169
6.44 2.1206 2.1162 0.0044 -14.0823 11.7562 -8.764 -2.3016
6.48 3.2265 3.1766 0.0499 -14.0453 15.4954 -5.2548 -1.3159
6.52 4.266 4.178 0.088 -13.9719 18.9971 -1.8575 -0.3683
6.56 5.2349 5.1156 0.1193 -13.8612 22.2461 1.4047 0.5359
6.6 6.1298 5.9855 0.1443 -13.7122 25.2288 4.5152 1.3931
6.64 6.9479 6.7839 0.164 -13.5239 27.934 7.4603 2.2004
6.68 7.6872 7.508 0.1792 -13.2958 30.3524 10.2281 2.9557
6.72 8.3457 8.1551 0.1906 -13.0273 32.4767 12.8086 3.6569
6.76 8.9223 8.7234 0.1989 -12.7179 34.3021 15.1929 4.3024
6.8 9.4166 9.2114 0.2052 -12.3675 35.8256 17.3742 4.8913
6.84 9.8284 9.6185 0.2099 -11.9758 37.0468 19.3476 5.4227
6.88 10.1578 9.9445 0.2133 -11.5429 37.9671 21.1099 5.8964
6.92 10.406 10.1898 0.2162 -11.0691 38.5904 22.659 6.3122
6.96 10.5742 10.3553 0.2189 -10.5546 38.9224 23.9951 6.6706
7 10.6642 10.4428 0.2214 -10 38.9711 25.1204 6.9723
7.04 10.6784 10.4542 0.2242 -9.4059 38.7465 26.0376 7.2184
7.08 10.6194 10.3924 0.227 -8.7732 38.2604 26.7523 7.4102
7.12 10.4905 10.2604 0.2301 -8.1028 37.5265 27.2708 7.5496
7.16 10.2952 10.062 0.2332 -7.3959 36.5601 27.6011 7.6386
7.2 10.0375 9.8013 0.2362 -6.6537 35.3782 27.7527 7.6796
7.24 9.7218 9.4829 0.2389 -5.8776 33.9993 27.7362 7.6753
7.28 9.3527 9.1116 0.2411 -5.0692 32.4431 27.5635 7.6286
7.32 8.9355 8.6929 0.2426 -4.2303 30.7305 27.2475 7.5428
7.36 8.4754 8.2324 0.243 -3.3626 28.8835 26.8021 7.4213
7.4 7.9781 7.736 0.2421 -2.4682 26.9248 26.2419 7.2678
7.44 7.4494 7.2097 0.2397 -1.549 24.8778 25.5825 7.0863
7.48 6.8955 6.6599 0.2356 -0.6074 22.7664 24.8397 6.8809
7.52 6.3227 6.093 0.2297 0.3543 20.615 24.0299 6.6558
7.56 5.7374 5.5155 0.2219 1.3338 18.4477 23.1698 6.4155
7.6 5.1461 4.934 0.2121 2.3285 16.2889 22.2764 6.1644
7.64 4.5554 4.3549 0.2005 3.3359 14.1628 21.3663 5.9071
7.68 3.9718 3.7847 0.1871 4.3531 12.0928 20.4566 5.6484
7.72 3.4019 3.2298 0.1721 5.3776 10.1021 19.5635 5.3927
7.76 2.8521 2.6963 0.1558 6.4065 8.2129 18.7034 5.1447
7.8 2.3288 2.1902 0.1386 7.4368 6.4466 17.892 4.9091

58
7.84 1.8379 1.7172 0.1207 8.4659 4.8234 17.1443 4.6901
7.88 1.3856 1.2828 0.1028 9.4906 3.3624 16.4749 4.4922
7.92 0.9771 0.8919 0.0852 10.508 2.0811 15.8975 4.3194
7.96 0.6178 0.5494 0.0684 11.5152 0.9956 15.4247 4.1757
8 0.3124 0.2596 0.0528 12.5093 0.1204 15.0683 4.0646
8.04 0.0651 0.0262 0.0389 13.4872 -0.5318 14.839 3.9896
8.08 -0.12 -0.1473 0.0273 14.4461 -0.9502 14.7463 3.9535
8.12 -0.2398 -0.2581 0.0183 15.3831 -1.1259 14.7983 3.9589
8.16 -0.2915 -0.3037 0.0122 16.2952 -1.052 15.002 4.0082
8.2 -0.2733 -0.2823 0.009 17.1797 -0.7237 15.3639 4.1031
8.24 -0.1838 -0.1927 0.0089 18.0339 -0.1382 15.8884 4.2447
8.28 -0.0221 -0.0341 0.012 18.855 0.705 16.5779 4.4338
8.32 0.212 0.1936 0.0184 19.6405 1.8045 17.433 4.6707
8.36 0.5179 0.4899 0.028 20.3879 3.1565 18.4537 4.9554
8.4 0.8943 0.854 0.0403 21.0948 4.7553 19.6392 5.2871
8.44 1.3393 1.2841 0.0552 21.7588 6.593 20.986 5.6645
8.48 1.8502 1.7783 0.0719 22.3779 8.6597 22.4894 6.0861
8.52 2.4241 2.3338 0.0903 22.9499 10.9435 24.143 6.5494
8.56 3.057 2.9475 0.1095 23.473 13.4308 25.9392 7.052
8.6 3.7447 3.6157 0.129 23.9454 16.106 27.8687 7.5906
8.64 4.4825 4.3343 0.1482 24.3654 18.9519 29.9206 8.1618
8.68 5.265 5.0985 0.1665 24.7315 21.9497 32.0829 8.7616
8.72 6.0866 5.9034 0.1832 25.0425 25.0793 34.3425 9.386
8.76 6.9415 6.7436 0.1979 25.2971 28.3191 36.6845 10.0303
8.8 7.8233 7.6133 0.21 25.4945 31.6465 39.0932 10.6899
8.84 8.7254 8.5064 0.219 25.6337 35.0379 41.5519 11.3598
8.88 9.6411 9.4167 0.2244 25.7141 38.4689 44.0427 12.0349
8.92 10.5639 10.3375 0.2264 25.7352 41.9144 46.5466 12.7098
8.96 11.4867 11.2621 0.2246 25.6969 45.3489 49.0453 13.3795
9 12.4027 12.1837 0.219 25.5988 48.7464 51.5186 14.0385
9.04 13.3052 13.0953 0.2099 25.4413 52.0811 53.9462 14.6815
9.08 14.187 13.99 0.197 25.2244 55.3271 56.3086 15.3035
9.12 15.0419 14.8608 0.1811 24.9486 58.4589 58.5842 15.8989
9.16 15.8635 15.7009 0.1626 24.6145 61.4512 60.7535 16.463
9.2 16.6454 16.5035 0.1419 24.223 64.2795 62.7969 16.991
9.24 17.3815 17.262 0.1195 23.775 66.9203 64.6946 17.4782
9.28 18.0661 17.9701 0.096 23.2716 69.3506 66.4273 17.92
9.32 18.6939 18.6216 0.0723 22.7142 71.549 67.9768 18.3122
9.36 19.2597 19.2108 0.0489 22.1042 73.4951 69.3259 18.6509
9.4 19.7588 19.7322 0.0266 21.4433 75.1702 70.4585 18.9325
9.44 20.1868 20.1807 0.0061 20.7333 76.5569 71.3596 19.1536
9.48 20.5398 20.5517 -0.0119 19.9761 77.6396 72.0155 19.3113
9.52 20.8142 20.841 -0.0268 19.1737 78.4048 72.4137 19.4027
9.56 21.0067 21.045 -0.0383 18.3284 78.8406 72.5436 19.4255
9.6 21.1147 21.1605 -0.0458 17.4426 78.9373 72.3959 19.3777
9.64 21.1358 21.1848 -0.049 16.5186 78.6872 71.9631 19.2576
9.68 21.0683 21.1159 -0.0476 15.5592 78.0848 71.2396 19.0639
9.72 20.9108 20.9524 -0.0416 14.5668 77.1269 70.2219 18.7957
9.76 20.6625 20.6934 -0.0309 13.5444 75.8125 68.9082 18.4526
9.8 20.3226 20.3386 -0.016 12.4947 74.1427 67.2986 18.0344

59
 9.84 19.8913 19.8883 0.003 11.4208 72.1212 65.3945 17.5411
9.88 19.3693 19.3435 0.0258 10.3254 69.7536 63.1998 16.9734
9.92 18.7572 18.7058 0.0514 9.2118 67.048 60.7208 16.3322
9.96 18.0567 17.9774 0.0793 8.083 64.0145 57.9648 15.619
10 17.2695 17.1609 0.1086 6.9421 60.6653 54.9413 14.8354
10.04 16.3982 16.2597 0.1385 5.7923 57.0149 51.6615 13.9835
10.08 15.4454 15.2777 0.1677 4.6367 53.0793 48.1384 13.0657
10.12 14.4145 14.2193 0.1952 3.4784 48.8769 44.3864 12.0848
10.16 13.3092 13.0895 0.2197 2.3206 44.4273 40.4217 11.044
10.2 12.1337 11.8935 0.2402 1.1664 39.7522 36.2614 9.9468
10.24 10.8925 10.6372 0.2553 0.019 34.8744 31.9245 8.7971
10.28 9.5908 9.3268 0.264 -1.1187 29.8181 27.4303 7.5989
10.32 8.2338 7.9688 0.265 -2.2437 24.6089 22.8 6.3567
10.36 6.8275 6.57 0.2575 -3.3529 19.2731 18.0552 5.0755
10.4 5.378 5.1376 0.2404 -4.4436 13.838 13.2182 3.7601
10.44 3.8919 3.6789 0.213 -5.513 8.3315 8.3119 2.4161
10.48 2.3758 2.2014 0.1744 -6.5583 2.782 3.3598 1.049
10.52 0.837 0.7128 0.1242 -7.5769 -2.782 -1.6151 -0.3355
10.56 -0.7166 -0.7794 0.0628 -8.5664 -8.3315 -6.5889 -1.7313
10.6 -2.2695 -2.2673 -0.0022 -9.5244 -13.838 -11.5475 -3.1318
10.64 -3.8039 -3.7434 -0.0605 -10.4486 -19.2731 -16.5222 -4.5257
10.68 -5.3086 -5.2003 -0.1083 -11.3369 -24.6089 -21.4654 -5.8991
10.72 -6.7761 -6.6307 -0.1454 -12.1873 -29.8181 -26.3347 -7.2414
10.76 -8.1991 -8.0271 -0.172 -12.998 -34.8744 -31.102 -8.5456
10.8 -9.5711 -9.3826 -0.1885 -13.7672 -39.7522 -35.7422 -9.8056
10.84 -10.8862 -10.6902 -0.196 -14.4933 -44.4273 -40.2315 -11.0159
10.88 -12.1388 -11.9435 -0.1953 -15.1751 -48.8769 -44.5472 -12.1716
10.92 -13.3238 -13.1363 -0.1875 -15.8111 -53.0793 -48.6683 -13.268
10.96 -14.4367 -14.2629 -0.1738 -16.4005 -57.0149 -52.5747 -14.3008
11 -15.4734 -15.3178 -0.1556 -16.9421 -60.6653 -56.2479 -15.2664
11.04 -16.43 -16.2962 -0.1338 -17.4354 -64.0145 -59.6709 -16.1614
11.08 -17.3035 -17.1938 -0.1097 -17.8797 -67.048 -62.828 -16.9827
11.12 -18.0912 -18.0066 -0.0846 -18.2745 -69.7536 -65.7055 -17.7278
11.16 -18.7908 -18.7314 -0.0594 -18.6197 -72.1212 -68.2913 -18.3945
11.2 -19.4005 -19.3655 -0.035 -18.9151 -74.1427 -70.5753 -18.9811
11.24 -19.9191 -19.9066 -0.0125 -19.1609 -75.8125 -72.5492 -19.4863
11.28 -20.346 -20.3533 0.0073 -19.3573 -77.1269 -74.2069 -19.9093
11.32 -20.6807 -20.7046 0.0239 -19.5047 -78.0848 -75.5442 -20.2495
11.36 -20.9236 -20.9601 0.0365 -19.6036 -78.6872 -76.559 -20.507
11.4 -21.0752 -21.1201 0.0449 -19.6549 -78.9373 -77.2513 -20.6821
11.44 -21.1367 -21.1856 0.0489 -19.6593 -78.8406 -77.623 -20.7755
11.48 -21.1096 -21.1579 0.0483 -19.6179 -78.4048 -77.6779 -20.7884
11.52 -20.9959 -21.0392 0.0433 -19.5318 -77.6396 -77.4218 -20.7224
11.56 -20.798 -20.832 0.034 -19.4024 -76.5569 -76.8623 -20.5792
11.6 -20.519 -20.5396 0.0206 -19.231 -75.1702 -76.0089 -20.3612
11.64 -20.1621 -20.1656 0.0035 -19.0192 -73.4951 -74.8733 -20.0711
11.68 -19.731 -19.7142 -0.0168 -18.7687 -71.549 -73.4685 -19.712
11.72 -19.2296 -19.19 -0.0396 -18.4812 -69.3506 -71.8091 -19.2871
11.76 -18.6625 -18.5983 -0.0642 -18.1585 -66.9203 -69.9105 -18.8
11.8 -18.0346 -17.9443 -0.0903 -17.8026 -64.2795 -67.7902 -18.2548

60
11.84 -17.3508 -17.2339 -0.1169 -17.4156 -61.4512 -65.4667 -17.6556
11.88 -16.6165 -16.4733 -0.1432 -16.9995 -58.4589 -62.9595 -17.007
11.92 -15.8376 -15.6688 -0.1688 -16.5565 -55.3271 -60.2885 -16.3138
11.96 -15.0198 -14.827 -0.1928 -16.0889 -52.0811 -57.4752 -15.5809
12 -14.1693 -13.9547 -0.2146 -15.5988 -48.7464 -54.5409 -14.8136
12.04 -13.2925 -13.0588 -0.2337 -15.0887 -45.3489 -51.5077 -14.0171
12.08 -12.3958 -12.1462 -0.2496 -14.5608 -41.9144 -48.3979 -13.1971
12.12 -11.4858 -11.224 -0.2618 -14.0174 -38.4689 -45.2339 -12.3592
12.16 -10.5693 -10.299 -0.2703 -13.4609 -35.0379 -42.0381 -11.509
12.2 -9.6528 -9.3782 -0.2746 -12.8937 -31.6465 -38.8326 -10.6524
12.24 -8.743 -8.4682 -0.2748 -12.3181 -28.3191 -35.6392 -9.7952
12.28 -7.8466 -7.5755 -0.2711 -11.7364 -25.0793 -32.4793 -8.9432
12.32 -6.9701 -6.7066 -0.2635 -11.1509 -21.9497 -29.3736 -8.102
12.36 -6.1197 -5.8675 -0.2522 -10.5638 -18.9519 -26.3419 -7.2774
12.4 -5.3017 -5.064 -0.2377 -9.9773 -16.106 -23.4035 -6.4747
12.44 -4.5219 -4.3014 -0.2205 -9.3936 -13.4308 -20.5764 -5.6995
12.48 -3.7859 -3.5848 -0.2011 -8.8148 -10.9435 -17.8778 -4.9566
12.52 -3.0988 -2.9187 -0.1801 -8.2427 -8.6597 -15.3236 -4.2511
12.56 -2.4653 -2.3073 -0.158 -7.6794 -6.593 -12.9281 -3.5873
12.6 -1.8899 -1.7544 -0.1355 -7.1267 -4.7553 -10.7046 -2.9695
12.64 -1.3763 -1.2629 -0.1134 -6.5863 -3.1565 -8.665 -2.4014
12.68 -0.9277 -0.8357 -0.092 -6.0599 -1.8045 -6.8195 -1.8865
12.72 -0.5468 -0.4746 -0.0722 -5.5489 -0.705 -5.1769 -1.4276
12.76 -0.2357 -0.1814 -0.0543 -5.0548 0.1382 -3.7443 -1.0271
12.8 0.0042 0.0431 -0.0389 -4.579 0.7237 -2.5272 -0.6871
12.84 0.1721 0.1985 -0.0264 -4.1225 1.052 -1.5293 -0.4087
12.88 0.2679 0.2849 -0.017 -3.6864 1.1259 -0.7525 -0.193
12.92 0.2922 0.3031 -0.0109 -3.2716 0.9502 -0.1976 -0.0402
12.96 0.2464 0.2544 -0.008 -2.8791 0.5318 0.1361 0.0501
13 0.1325 0.1406 -0.0081 -2.5093 -0.1204 0.2518 0.0788
13.04 -0.0472 -0.0357 -0.0115 -2.1629 -0.9956 0.1549 0.0473
13.08 -0.2897 -0.2718 -0.0179 -1.8402 -2.0811 -0.1477 -0.0424
13.12 -0.5911 -0.5641 -0.027 -1.5415 -3.3624 -0.6485 -0.188
13.16 -0.9472 -0.9089 -0.0383 -1.2669 -4.8234 -1.3383 -0.3866
13.2 -1.3533 -1.3017 -0.0516 -1.0165 -6.4466 -2.2057 -0.6345
13.24 -1.8042 -1.738 -0.0662 -0.79 -8.2129 -3.238 -0.9281
13.28 -2.2944 -2.2124 -0.082 -0.5871 -10.1021 -4.4212 -1.2631
13.32 -2.8178 -2.7198 -0.098 -0.4076 -12.0928 -5.7401 -1.6348
13.36 -3.3685 -3.2542 -0.1143 -0.2509 -14.1628 -7.178 -2.0384
13.4 -3.9399 -3.8098 -0.1301 -0.1162 -16.2889 -8.7174 -2.4688
13.44 -4.5255 -4.3804 -0.1451 -0.0029 -18.4477 -10.3398 -2.9207
13.48 -5.1188 -4.9596 -0.1592 0.0899 -20.615 -12.0259 -3.3884
13.52 -5.7129 -5.541 -0.1719 0.1632 -22.7664 -13.7559 -3.8665
13.56 -6.3013 -6.118 -0.1833 0.2181 -24.8778 -15.5092 -4.3493
13.6 -6.8771 -6.6841 -0.193 0.2559 -26.9248 -17.2651 -4.831
13.64 -7.434 -7.2329 -0.2011 0.2776 -28.8835 -19.0023 -5.306
13.68 -7.9655 -7.7579 -0.2076 0.2848 -30.7305 -20.7003 -5.7688
13.72 -8.4655 -8.2528 -0.2127 0.2788 -32.4431 -22.338 -6.2137
13.76 -8.928 -8.7116 -0.2164 0.2611 -33.9993 -23.8949 -6.6356
13.8 -9.3474 -9.1283 -0.2191 0.2333 -35.3782 -25.3509 -7.0291

61
13.84 -9.7181 -9.4973 -0.2208 0.1969 -36.5601 -26.6868 -7.3893
13.88 -10.0353 -9.8134 -0.2219 0.1538 -37.5265 -27.8839 -7.7114
13.92 -10.294 -10.0715 -0.2225 0.1054 -38.2604 -28.9247 -7.9911
13.96 -10.4901 -10.2672 -0.2229 0.0536 -38.7465 -29.7922 -8.2239
14 -10.6197 -10.3963 -0.2234 0 -38.9711 -30.4714 -8.4061
14.04 -10.6791 -10.4552 -0.2239 -0.0536 -38.9224 -30.9486 -8.5341
14.08 -10.6654 -10.4406 -0.2248 -0.1054 -38.5904 -31.2108 -8.6046
14.12 -10.5759 -10.3499 -0.226 -0.1538 -37.9671 -31.2475 -8.6149
14.16 -10.4086 -10.181 -0.2276 -0.1969 -37.0468 -31.0491 -8.5622
14.2 -10.1616 -9.9323 -0.2293 -0.2333 -35.8256 -30.6084 -8.4447
14.24 -9.8338 -9.6029 -0.2309 -0.2611 -34.3021 -29.9195 -8.2604
14.28 -9.4245 -9.1923 -0.2322 -0.2788 -32.4767 -28.9784 -8.0081
14.32 -8.9335 -8.7007 -0.2328 -0.2848 -30.3524 -27.7831 -7.6868
14.36 -8.3611 -8.1289 -0.2322 -0.2776 -27.934 -26.3336 -7.296
14.4 -7.708 -7.4783 -0.2297 -0.2559 -25.2288 -24.6316 -6.8357
14.44 -6.9754 -6.7508 -0.2246 -0.2181 -22.2461 -22.6807 -6.3061
14.48 -6.1654 -5.9489 -0.2165 -0.1632 -18.9971 -20.4862 -5.708
14.52 -5.2798 -5.0757 -0.2041 -0.0899 -15.4954 -18.0557 -5.0426
14.56 -4.3218 -4.1349 -0.1869 0.0029 -11.7562 -15.3979 -4.3115
14.6 -3.2945 -3.1305 -0.164 0.1162 -7.7966 -12.5237 -3.5166
14.64 -2.2017 -2.0672 -0.1345 0.2509 -3.6354 -9.4454 -2.6605
14.68 -1.0475 -0.95 -0.0975 0.4076 0.7067 -6.1768 -1.746
14.72 0.1634 0.2154 -0.052 0.5871 5.208 -2.733 -0.7764
14.76 1.4227 1.4233 -0.0006 0.79 9.8452 0.8701 0.2447
14.8 2.7171 2.6674 0.0497 1.0165 14.5938 4.6479 1.3114
14.84 4.0362 3.9412 0.095 1.2669 19.4284 8.5859 2.4139
14.88 5.3715 5.2381 0.1334 1.5415 24.3226 12.6501 3.5432
14.92 6.7154 6.551 0.1644 1.8402 29.2496 16.8156 4.692
14.96 8.0601 7.8726 0.1875 2.1629 34.1818 21.0602 5.854
15 9.3983 9.1957 0.2026 2.5093 39.0916 25.3603 7.0227
15.04 10.7218 10.5127 0.2091 2.8791 43.9509 29.6934 8.1919
15.08 12.0238 11.8163 0.2075 3.2716 48.7321 34.034 9.3551
15.12 13.2975 13.0989 0.1986 3.6864 53.4074 38.3584 10.5062
15.16 14.5357 14.353 0.1827 4.1225 57.9499 42.6433 11.6395
15.2 15.7316 15.5715 0.1601 4.579 62.3329 46.8641 12.7491
15.24 16.879 16.7473 0.1317 5.0548 66.5308 50.9958 13.8291
15.28 17.9719 17.8733 0.0986 5.5489 70.5188 55.0146 14.8739
15.32 19.0049 18.943 0.0619 6.0599 74.2734 58.8977 15.8785
15.36 19.9725 19.9499 0.0226 6.5863 77.7722 62.6228 16.8379
15.4 20.8698 20.8882 -0.0184 7.1267 80.9943 66.1682 17.7474
15.44 21.6924 21.7522 -0.0598 7.6794 83.9207 69.513 18.6025
15.48 22.4362 22.5367 -0.1005 8.2427 86.5336 72.6378 19.3991
15.52 23.0974 23.237 -0.1396 8.8148 88.8174 75.5244 20.1334
15.56 23.6729 23.8489 -0.176 9.3936 90.7585 78.1561 20.8019
15.6 24.16 24.3687 -0.2087 9.9773 92.345 80.5177 21.4014
15.64 24.5562 24.7933 -0.2371 10.5638 93.5675 82.5958 21.9292
15.68 24.8599 25.12 -0.2601 11.1509 94.4186 84.3785 22.3829
15.72 25.0695 25.347 -0.2775 11.7364 94.8931 85.8559 22.7603
15.76 25.1842 25.4727 -0.2885 12.3181 94.9881 87.0201 23.0598
15.8 25.2035 25.4965 -0.293 12.8937 94.7031 87.8649 23.28

62
15.84 25.1274 25.4182 -0.2908 13.4609 94.0399 88.3863 23.42
15.88 24.9564 25.2383 -0.2819 14.0174 93.0023 88.5819 23.4793
15.92 24.6914 24.9577 -0.2663 14.5608 91.5967 88.4518 23.4575
15.96 24.3338 24.5784 -0.2446 15.0887 89.8316 87.9978 23.3549
16 23.8852 24.1024 -0.2172 15.5988 87.7175 87.2234 23.172
16.04 23.348 23.5327 -0.1847 16.0889 85.2673 86.1343 22.9098
16.08 22.7249 22.8728 -0.1479 16.5565 82.4956 84.7376 22.5693
16.12 22.0191 22.1265 -0.1074 16.9995 79.4191 83.0427 22.1523
16.16 21.2341 21.2985 -0.0644 17.4156 76.0561 81.061 21.661
16.2 20.3737 20.3935 -0.0198 17.8026 72.4267 78.8049 21.0976
16.24 19.4425 19.4172 0.0253 18.1585 68.5526 76.288 20.4648
16.28 18.445 18.3752 0.0698 18.4812 64.4566 73.5261 19.7658
16.32 17.3864 17.2738 0.1126 18.7687 60.163 70.5358 19.0038
16.36 16.2721 16.1195 0.1526 19.0192 55.697 67.3349 18.1826
16.4 15.1078 14.9191 0.1887 19.231 51.0847 63.9422 17.3063
16.44 13.8995 13.6798 0.2197 19.4024 46.353 60.3775 16.379
16.48 12.6537 12.4087 0.245 19.5318 41.5293 56.6609 15.4055
16.52 11.3768 11.1131 0.2637 19.6179 36.6413 52.8136 14.3905
16.56 10.0758 9.8007 0.2751 19.6593 31.7168 48.8569 13.339
16.6 8.7574 8.4787 0.2787 19.6549 26.7837 44.8125 12.2566
16.64 7.4287 7.1548 0.2739 19.6036 21.8697 40.702 11.1484
16.68 6.0971 5.8363 0.2608 19.5047 17.0019 36.5471 10.0202
16.72 4.77 4.5305 0.2395 19.3573 12.2071 32.3692 8.8778
16.76 3.4543 3.2446 0.2097 19.1609 7.5111 28.1902 7.727
16.8 2.1575 1.9855 0.172 18.9151 2.9389 24.03 6.5737
16.84 0.8868 0.76 0.1268 18.6197 -1.4856 19.9092 5.4238
16.88 -0.3507 -0.4256 0.0749 18.2745 -5.7399 15.8474 4.2832
16.92 -1.5444 -1.5653 0.0209 17.8797 -9.8028 11.8612 3.1579
16.96 -2.6815 -2.6535 -0.028 17.4354 -13.6544 7.936 2.0559
17 -3.7549 -3.685 -0.0699 16.9421 -17.277 4.1012 0.9872
17.04 -4.7595 -4.655 -0.1045 16.4005 -20.6541 0.3889 -0.0411
17.08 -5.6918 -5.5592 -0.1326 15.8111 -23.7715 -3.1802 -1.0243
17.12 -6.5486 -6.3938 -0.1548 15.1751 -26.6167 -6.5896 -1.9587
17.16 -7.3275 -7.1554 -0.1721 14.4933 -29.1795 -9.8252 -2.8414
17.2 -8.0266 -7.8413 -0.1853 13.7672 -31.4516 -12.8745 -3.6701
17.24 -8.6443 -8.4492 -0.1951 12.998 -33.427 -15.7267 -4.4425
17.28 -9.1798 -8.9775 -0.2023 12.1873 -35.1017 -18.3722 -5.1568
17.32 -9.6328 -9.4251 -0.2077 11.3369 -36.474 -20.8039 -5.8118
17.36 -10.0033 -9.7916 -0.2117 10.4486 -37.5444 -23.0165 -6.4066
17.4 -10.292 -10.0772 -0.2148 9.5244 -38.3156 -25.0058 -6.9407
17.44 -10.5 -10.2824 -0.2176 8.5664 -38.7923 -26.7694 -7.4138
17.48 -10.6288 -10.4087 -0.2201 7.5769 -38.9816 -28.3076 -7.8262
17.52 -10.6806 -10.4579 -0.2227 6.5583 -38.8923 -29.6214 -8.1785
17.56 -10.6579 -10.4323 -0.2256 5.513 -38.5353 -30.7138 -8.4715
17.6 -10.5635 -10.335 -0.2285 4.4436 -37.9235 -31.5898 -8.7066
17.64 -10.4009 -10.1693 -0.2316 3.3529 -37.0713 -32.2553 -8.8855
17.68 -10.1739 -9.9392 -0.2347 2.2437 -35.9949 -32.7184 -9.01
17.72 -9.8866 -9.649 -0.2376 1.1187 -34.7122 -32.9881 -9.0825
17.76 -9.5436 -9.3035 -0.2401 -0.019 -33.2421 -33.0749 -9.1057
17.8 -9.1498 -8.9079 -0.2419 -1.1664 -31.605 -32.9904 -9.0823

63
17.84 -8.7105 -8.4675 -0.243 -2.3206 -29.8224 -32.7476 -9.0156
17.88 -8.231 -7.9883 -0.2427 -3.4784 -27.9166 -32.36 -8.909
17.92 -7.7173 -7.4762 -0.2411 -4.6367 -25.9108 -31.8424 -8.7662
17.96 -7.1752 -6.9373 -0.2379 -5.7923 -23.8286 -31.21 -8.5912
18 -6.6111 -6.3782 -0.2329 -6.9421 -21.6942 -30.4787 -8.388
18.04 -6.0312 -5.8052 -0.226 -8.083 -19.5318 -29.665 -8.161
18.08 -5.4421 -5.2248 -0.2173 -9.2118 -17.3657 -28.7854 -7.9145
18.12 -4.8502 -4.6437 -0.2065 -10.3254 -15.2203 -27.8571 -7.6532
18.16 -4.2623 -4.0683 -0.194 -11.4208 -13.1193 -26.8968 -7.3817
18.2 -3.6847 -3.5049 -0.1798 -12.4947 -11.0861 -25.9213 -7.1046
18.24 -3.1241 -2.96 -0.1641 -13.5444 -9.1434 -24.9474 -6.8267
18.28 -2.5867 -2.4394 -0.1473 -14.5668 -7.3131 -23.9914 -6.5526
18.32 -2.0789 -1.9492 -0.1297 -15.5592 -5.6159 -23.0693 -6.287
18.36 -1.6066 -1.4948 -0.1118 -16.5186 -4.0715 -22.1962 -6.0345
18.4 -1.1755 -1.0816 -0.0939 -17.4426 -2.6982 -21.3871 -5.7993
18.44 -0.791 -0.7144 -0.0766 -18.3284 -1.5129 -20.6559 -5.5858
18.48 -0.4581 -0.3977 -0.0604 -19.1737 -0.5309 -20.0155 -5.3978
18.52 -0.1812 -0.1356 -0.0456 -19.9761 0.2343 -19.4782 -5.2392
18.56 0.0354 0.0682 -0.0328 -20.7333 0.7708 -19.0548 -5.1133
18.6 0.1883 0.2107 -0.0224 -21.4433 1.0689 -18.7556 -5.0231
18.64 0.2743 0.2891 -0.0148 -22.1042 1.1205 -18.5887 -4.9715
18.68 0.2913 0.3015 -0.0102 -22.7142 0.9199 -18.5623 -4.9608
18.72 0.2375 0.2461 -0.0086 -23.2716 0.4632 -18.6832 -4.9926
18.76 0.112 0.122 -0.01 -23.775 -0.2512 -18.956 -5.0684
18.8 -0.0859 -0.0711 -0.0148 -24.223 -1.2229 -19.3834 -5.189
18.84 -0.356 -0.3332 -0.0228 -24.6145 -2.4492 -19.967 -5.3549
18.88 -0.6974 -0.6636 -0.0338 -24.9486 -3.9255 -20.708 -5.5661
18.92 -1.1084 -1.0609 -0.0475 -25.2244 -5.6448 -21.6055 -5.8219
18.96 -1.5867 -1.5234 -0.0633 -25.4413 -7.5984 -22.6571 -6.1214
19 -2.1295 -2.0486 -0.0809 -25.5988 -9.7753 -23.8584 -6.4629
19.04 -2.7335 -2.6336 -0.0999 -25.6969 -12.1626 -25.2043 -6.8444
19.08 -3.3943 -3.2751 -0.1192 -25.7352 -14.7459 -26.6876 -7.2633
19.12 -4.1077 -3.969 -0.1387 -25.7141 -17.5087 -28.2996 -7.7168
19.16 -4.8685 -4.711 -0.1575 -25.6337 -20.433 -30.0305 -8.2016
19.2 -5.6713 -5.4962 -0.1751 -25.4945 -23.4993 -31.8693 -8.7141
19.24 -6.5103 -6.3195 -0.1908 -25.2971 -26.6868 -33.8032 -9.2502
19.28 -7.3794 -7.1751 -0.2043 -25.0425 -29.9733 -35.8186 -9.8057
19.32 -8.2722 -8.0573 -0.2149 -24.7315 -33.3357 -37.9005 -10.3763
19.36 -9.182 -8.9598 -0.2222 -24.3654 -36.75 -40.0333 -10.9572
19.4 -10.1021 -9.8762 -0.2259 -23.9454 -40.1915 -42.1996 -11.5435
19.44 -11.0256 -10.7997 -0.2259 -23.473 -43.6346 -44.382 -12.1304
19.48 -11.9459 -11.7237 -0.2222 -22.9499 -47.0539 -46.5625 -12.7128
19.52 -12.856 -12.6412 -0.2148 -22.3779 -50.4232 -48.7225 -13.286
19.56 -13.749 -13.5452 -0.2038 -21.7588 -53.7168 -50.8429 -13.8449
19.6 -14.6182 -14.4288 -0.1894 -21.0948 -56.9089 -52.9043 -14.3845
19.64 -15.4572 -15.2851 -0.1721 -20.3879 -59.974 -54.8869 -14.9
19.68 -16.2598 -16.1073 -0.1525 -19.6405 -62.8873 -56.7724 -15.3868
19.72 -17.0195 -16.8886 -0.1309 -18.855 -65.6248 -58.5419 -15.8404
19.76 -17.7306 -17.6227 -0.1079 -18.0339 -68.1631 -60.1765 -16.2564
19.8 -18.3874 -18.3033 -0.0841 -17.1797 -70.4801 -61.6581 -16.6306

64
19.84 -18.9848 -18.9243 -0.0605 -16.2952 -72.5548 -62.9699 -16.9592
19.88 -19.5178 -19.4803 -0.0375 -15.3831 -74.3677 -64.0956 -17.2384
19.92 -19.982 -19.9658 -0.0162 -14.4461 -75.9006 -65.0202 -17.4649
19.96 -20.373 -20.3761 0.0031 -13.4872 -77.1371 -65.7295 -17.6357
20 -20.687 -20.7068 0.0198 -12.5093 -78.0627

65
9.2 PLOT OF TRAINED ANN FROM SYSTEM IDENTIFICATION CODING:

Figure 24: Trend of root-mean-squared (RMS) error for the entire training set over several training iterations

66
 Chapter 10

CONCLUSION

67
10.1 DISCUSSIONS:

The speed trajectory of a dc motor has been successfully controlled by using a neural network

based model reference adaptive control (MRAC) technique. A multi-layer perceptron is used

to identify the motor dynamics. The trained neural network is then used to derive the

appropriate control voltage for trajectory control of the rotor speed.

The controller design does not require explicit knowledge of the motor/load dynamics which

is a useful feature when dealing with parameter and load uncertainties. The actual tuning of

the controller only involves minimal effort. The motor displays good tracking performance

for different speed trajectories. The robustness of the controller is positively established by

evaluating its performance under noisy loading conditions. The controller remains stable for a

wide range of sampling frequencies.

It is believed that better control performance can be obtained by training the neural network

on-line. However, the resulting increased computational overhead at each sampling instant

calls for a faster and more efficient software/hardware arrangement.

The unknown, time invariant, nonlinear operating characteristics of the dc motor and its load

have been successfully captured by an ANN. The concepts of model reference adaptive

control have been used in conjunction with the trained ANN to achieve trajectory control

of the rotor speed.

10.2 FUTURE SCOPE:

The ability to deal with a noisy operating environment is always an important consideration

for electric drives control. Noise can be introduced due to several reasons. Two main causes

are motor-load parameter drift and quantization and resolution errors of the speed and/or

position encoders. A high performance drive controller should be robust enough to maintain

accurate tracking performance regardless of the noisy operating environment. ANN

68
based controllers have the inherent noise rejection capability of the ANN built into them.

Therefore ANN based controllers are expected to display superior performance under noisy

operating environments.

The investigation of the controller performance under noise can be a cruitial phenomenon

that can be studied through the necessary simulations.

69
 Chapter 11

BIBLIOGRAPHY

70
Antsaklis, P. J. (April, 1990). Neural Networks in control systems. IEEE control systems magazine,
volume-10 , 3-5.

Cajueiro, D. O., & Hemerly, E. M. (December, 2003). Direct Adaptive Control using Feedforward
Neural Networks. Revista Control & Automation, Vol.14, No.4 , 348-358.

Chen, S., Billings, S., & Grant, P. (April, 1990). Non-linear system identification using neural
networks. IEEE Transactions, volume-51 , 1191-1214.

DOURATSOS, I., & BARRY, J. (2007). NEURAL NETWORK BASED MODEL REFERENCE
ADAPTIVE. International journal of information and systems sciences, volume-3 , 161-179.

Dzung, P., & Phuong, L. (2002). ANN - Control System DC Motor. IEEE Transactions , 82-90.

El-Sharkawi, M. A., & Weerasooriya, S. (September, 1989). Adaptive tracking control for high
performance dc drives. IEEE Transactions on Energy Conversion, volume-5 , 122-128.

El-Sharkawi, M. A., & Weerasooriya, S. (December, 1991). Identification and control of a dc motor
using back-propagation neural networks. IEEE Transactions on energy conversion, volume-6 , 663-
669.

Hagan, M. T., Demuth, H. B., & Jesus, O. (2000). An Introduction To The Useof Neural Networks.
Electrical and Computer Engg. , 1-23.

Kotaru, R., & Rubai, A. (June, 2000). Online Identification and control of a dc motor using learning
adaptation of neural networks. IEEE Transactions on industry applications, volume-36 , 935-942.

Liu, Z., Zhuang, X., & Wang, S. (2003). Speed Control of a DC Motor using BP Neural Networks.
IEEE Transactions , 832-835.

Minkova, M., Rodgerson, J. L., & Harley, R. G. (1996). Current Limitations in the Adaptive Neural
Speed Control of a DC Motor. IEEE Transactions , 837-842.

Moleykutty, G. (2008). Speed Control of Separately Excited DC Motor. American Journal of Applied
Sciences 5 , 227-233.

Odry, P., & Kavai, G. (2001). Neural Motor Control. Polytechnical Engg. , 55-66.

Parthasarathy, K., & Narendra, K. (March, 1990). Identification and control of dynamical systems
using neural networks. IEEE Transactions on Neural Networks, volume-1 , 4-27.

Weerasooriya, S., & El-Sharkawi, M. A. (March 1993). Laboratory Implementation of a Neural

Network Trajectory Controller for a DC Motor. IEEE Transactions on Energy Conversions, Vol. 8,
No. 1 , 107-113.

Zilkova, J., Timko, J., & Girovsky, P. (2006). Nonlinear System Control Using Neural Networks.
Acta Polytechnica Hungarica, Vol.3; No.4 , 85-94.

71